Re: socket creation synchronization
Re: socket creation synchronization
- Subject: Re: socket creation synchronization
- From: Christopher Vitale <email@hidden>
- Date: Mon, 05 Dec 2005 18:53:36 -0500
- Thread-topic: socket creation synchronization
Hello,
I'm calling this:
socket_t so;
sock_socket(PF_SYSTEM, SOCK_DGRAM, SYSPROTO_CONTROL, NULL, NULL, &so);
This should be fine. It works almost all the time. ;-)
In my stack trace shows sock_socket calling socreate, which calls
pffindproto.
struct protosw *
pffindproto(family, protocol, type)
int family, protocol, type;
{
register struct protosw *pr;
lck_mtx_assert(domain_proto_mtx, LCK_MTX_ASSERT_NOTOWNED);
lck_mtx_lock(domain_proto_mtx);
pr = pffindproto_locked(family, protocol, type);
lck_mtx_unlock(domain_proto_mtx);
return (pr);
}
I die on the lck_mtx_assert, xnu-792.6.22/bsd/kern/uipc_domain.c:367. I'm
still in the generic socket handling code at this point. The fact that I'm
creating a SYSPROTO_CONTROL socket shouldn't matter. I think a userland
socket may be holding the domain_proto_mtx lock.
A remote employee is the only one that cause the crash consistently, I've
been relying on his panic.log. A two-machine debug remains elusive.
Thanks,
vitale
Subject: Re: socket creation synchronization
From: Josh Graessley <email@hidden>
Date: Mon, 5 Dec 2005 15:15:52 -0800
What type of socket are you creating? What gets passed in so sock_socket for
domain, type and protocol?
Adding to the list of protocols by registering a new protosw or domain
structure is not supported. The locking model has not been documented.
-josh
On 12/5/05 4:30 PM, "Christopher Vitale" <email@hidden>
wrote:
> Hello List,
>
> I have a question about socket creation in 10.4.3.
>
> You can arrive at socreate by calling the kpi sock_socket, or from the
> userland syscall socket. Regardless, socreate calls pffindproto or
> pffindtype. These domain functions both assert that nobody else is holding a
> certain lock. How is this guaranteed? There isn't even a funnel associated
> with the socket syscall in syscalls.master. I looked around the source for a
> while but nothing jumped out at me.
>
> I'm interested because I've got a kext that creates a connection via
> sock_socket. When I call sock_socket while other (userland) sockets are
> being created I'll occasionally die a horrible death on the pffindproto lock
> assert. I'll only create a single connection for the life of the kext. Does
> socket creation in the kernel bypass synchronization at a higher level?
>
> Also, the socket I create is a system socket to another kext. Is this an
> appropriate way for kexts to share information? I take the necessary
> precautions like not allowing the server kext to unload while connected,
> forcing the client kext to disconnect when unloaded, etc.
>
> Thanks,
>
> vitale
>
> _______________________________________________
> Do not post admin requests to the list. They will be ignored.
> Darwin-kernel mailing list (email@hidden)
> Help/Unsubscribe/Update your Subscription:
> ech.com
>
> This email sent to email@hidden
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Darwin-kernel mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden